Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hashkey collisions (typical numbers)

Author: Sune Fischer

Date: 08:19:12 04/07/04

Go up one level in this thread


On April 07, 2004 at 11:12:23, Renze Steenhuisen wrote:

>
>>As would I.  This has been tested over and over and over.  64 bit signatures
>>won't produce signature collisions every 7M nodes without a serious bug or very
>>poor random numbers...
>
>Alright, I agree...
>
>I am using TWO-DEEP as a replacement scheme.
>    (move the first two the second entry when new data has a higher draft,
>     else overwrite second)
>Not clearing the hashtable between iteration, just between succeeding searches.
>
>I checked all do_move's and undo_move's by comparing the incremental hashkey
>with the fresh-from-scratch generated hashkey before AND after every call,
>including NULL-move.

I hope it is not a stupid question, but have you also checked that your
fresh-from-scratch generator works fine?

Ie. you must represent the position uniquely so side to move, e.p and castle
status must also be stored in the key.

>I am using the same random number generator as Crafty is.
>
>I rechecked the values that are used:
>Overall:
>  ~50%  0's
>  ~50%  1's
>Per bit-position over all 368 keys (P=48, N=64, B=64, R=64, Q=64, K=64)
>all have
>  ~50%  0's
>  ~50%  1's

I'm sure they are fine, but good idea to print them out just once to have
a quick check with your own eyes.

-S.
>I am desperate for help!
>
>Renze



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.